Clearly, there is no CPU bottleneck. Combined with the Top 5 Timed Events report snippet shown in Figure 9-15, we can see there is an IO bottleneck.

Because it's obvious that there is an IO read bottleneck, the number of server process IO read requests should be a good unit of work. This is the instance statistic (v$sysstat) physical read IO requests. Oracle tracks physical IO by SQL statement, allowing our response-time mathematics, the response-time curve, our performance-improving strategy, and communication to be easily understood and well founded. If there were an IO write bottleneck, the instance statistic db block changes would be another good unit of work candidate.

The AWR Instance Statistics section showed the physical read IO requests (breads for short) statistic to be 148,439. Before we calculate the service and queue times, the arrival rate based on our unit of work needs to be calculated. Here is the arrival rate math: